Tan Qipeng, Zhang Tiandong, Wu Shaocheng, Gao Jiachen, Song Bin
School of Electrical Engineering and Automation, Wuhan University, Wuhan 430072, China.
Department of Electrical, Electronic, and Information Engineering, 'Guglielmo Marconi', University of Bologna, 40136 Bologna, Italy.
Sensors (Basel). 2022 Mar 20;22(6):2395. doi: 10.3390/s22062395.
Partial discharge (PD) is a common phenomenon of insulation aging in air-insulated switchgear and will change the gas composition in the equipment. However, it is still a challenge to diagnose and identify the defect types of PD. This paper conducts enclosed experiments based on gas sensors to obtain the concentration data of the characteristic gases CO, NO, and O under four typical defects. The random forest algorithm with grid search optimization is used for fault identification to explore a method of identifying defect types through gas concentration. The results show that the gases concentration variations do have statistical characteristics, and the RF algorithm can achieve high accuracy in prediction. The combination of a sensor and a machine learning algorithm provides the gas component analysis method a way to diagnose PD in an air-insulated switchgear.
局部放电（PD）是空气绝缘开关设备中绝缘老化的常见现象，会改变设备内的气体成分。然而，诊断和识别局部放电的缺陷类型仍然是一项挑战。本文基于气体传感器进行封闭实验，以获取四种典型缺陷下特征气体CO、NO和O的浓度数据。采用带网格搜索优化的随机森林算法进行故障识别，探索通过气体浓度识别缺陷类型的方法。结果表明，气体浓度变化确实具有统计特征，随机森林算法在预测中能达到较高精度。传感器与机器学习算法的结合为气体成分分析方法提供了一种诊断空气绝缘开关设备中局部放电的途径。
